Proceedings of the Tenth International AAAI Conference on
Web and Social Media (ICWSM 2016)
Are You Charlie or Ahmed?
Cultural Pluralism in Charlie Hebdo Response on Twitter
Jisun An, Haewoon Kwak,
Yelena Mejova∗
Qatar Computing Research Institute, Qatar
jan,hkwak,ymejova@qf.org.qa
Sonia Alonso Saenz De Oger
Braulio Gomez Fortes
Georgetown University School
of Foreign Service, Qatar
sa1197@georgetown.edu
Deusto University, Bilbao, Spain
braulio.gomez@deusto.es
cally, by representatives of radical right parties as a statement against the Islamization of Europe.
The objective of this paper is to understand the social factors that contribute to online individual behavior. In particular we use Charlie Hebdo as a case study of three prominent
sociological theories modeling attention and opinion, ranging in the assumptions about the formation of individual’s
opinion:
Abstract
We study the response to the Charlie Hebdo shootings of January 7, 2015 on Twitter across the globe. We ask whether the
stances on the issue of freedom of speech can be modeled using established sociological theories, including Huntington’s
culturalist Clash of Civilizations, and those taking into consideration social context, including Density and Interdependence theories. We find support for Huntington’s culturalist explanation, in that the established traditions and norms
of one’s “civilization” predetermine some of one’s opinion.
However, at an individual level, we also find social context to
play a significant role, with non-Arabs living in Arab countries using #JeSuisAhmed (“I am Ahmed”) five times more
often when they are embedded in a mixed Arab/non-Arab
(mention) network. Among Arabs living in the West, we find
a great variety of responses, not altogether associated with the
size of their expatriate community, suggesting other variables
to be at play.1
• Clash of civilizations – “the great divisions among humankind (...) will be cultural” (Huntington and others
1993). Individuals’ opinions and behavior are determined
by the culture in which they are socialized. Cultures,
on the other hand, organize around different civilizations,
such as the Western Christian and the Islamic civilizations.
• Density theory – individuals’ opinions are influenced by
the socio-demographic and/or cultural density of their offline social context. The amount of interaction between
Muslims and non-Muslims, both in the West and Middle East, as expat communities become integrated into the
cultural fabric of its host nation, may affect the opinion of
one about the other.
• Interdependence theory – individuals’ opinions are influenced by the structure of online interactions within
their social network. The personal connections the individuals have, including those online, may change the their
worldview.
Introduction
On the 7th of January 2015 the Paris offices of the French
satirical weekly newspaper Charlie Hebdo were assaulted by
two brothers, French citizens born to Algerian parents, who
killed 11 persons and injured 11 more. The brothers claimed
to belong to Al Qaeda’s branch in Yemen. Charlie Hebdo is
a controversial magazine, partly due to the paper’s highly
secularist, and even openly anti-religious, articles making
fun of Catholicism, Judaism and Islam. The terrorist attack
against Charlie Hebdo was therefore widely interpreted as
an attack against freedom of expression and freedom of the
press, core principles of liberal democratic societies.
The social media, and Twitter in particular, reacted immediately upon the attack. The hashtags #CharlieHebdo
and #JeSuisCharlie (“I am Charlie”) became an explicit
endorsement of freedom of expression and freedom of the
press and travelled fast and wide in Twitter. Qualifying
or directly opposing this endorsement, other hashtags soon
followed: #JeSuisAhmed (“I am Ahmed”) and #JeNeSuisPasCharlie (“I am not Charlie”). The latter was used not
only by people who were against the editorial line of Charlie Hebdo for being offensive to Islam but also, paradoxi-
The aim of this work is both to re-examine the above theories in the context of culturally-charged online discussion,
and to better understand the actors within the online phenomena of #JeSuisCharlie.
Modeling Opinion Formation
Scholars of political behavior have long demonstrated that
individual political behavior changes as a function of social context (Allardt and Pesonen 1967; Huckfeldt 2009a;
2009b; Przeworski 1974; Wright 1976). Studies of voting
behavior, for example, have shown that vote choice is not
the result of an individual decision taken in isolation from
the characteristics of the social context in which the individual is embedded. As Przeworski put several decades ago:
“In order to understand political behavior, it is necessary to
Copyright © 2016, Association for the Advancement of Artificial
Intelligence (www.aaai.org). All rights reserved.
1
* First three authors’ names are in alphabetical order.
2
treat individuals within the context of their social interactions” (Przeworski 1974).
Theories we chose are well-established in social science
community, and their use in big data analysis extends both
computational and social science fields.
We begin with a macro-scale, deterministic cultural explanation offered by Huntingtons civilizational theory. In
the Clash of Civilizations seminal paper (Huntington and
others 1993), Samuel Huntington argues that “the fault lines
between civilizations will be the battle lines of the future”. A
civilization is defined as a cultural entity, the highest among
humans, and the broadest level of cultural identity. Religion
is a major civilizational component. Two of the major civilizations discussed by Huntington are the Christian Western
civilization and the Islamic civilization. According to Huntington, the Islamic civilization is incompatible with democratic values such as freedom of speech and freedom of the
press. An individual’s opinion on the Charlie Hebdo attack
will therefore be determined by the civilization she belongs
to, irrespective of the offline social context and online structure of interactions:
[H1] Opinions expressed about the Charlie Hebdo shootings are divided along “civilizational” faultlines, with a
higher proportion of pro-free speech tweets by users in Western Christian civilization countries, and a higher one of proMuslim tweets by users in the Islamic civilization countries.
Clash of Civilizations theory has been previously tested
in international communications networks in both social
media and email by State et al. (State et al. 2015), who
conclude its continuing endurance: “a bottom-up analysis
confirms the persistence of the eight culturally differentiated
civilizations posited by Huntington, with the divisions
corresponding to differences in language, religion, economic development, and spatial distance”. Going beyond
communication volume, we test the Clash of Civilization
theory by examining individuals’ behavior as captured from
the usage of different hashtags.
jority feels threatened by increasing minority. Thus we pose
two hypotheses for Western users, reflecting the two alternatives (with mirror theories possible for Muslim users):
[H2a] The higher the proportion of Muslims in the population, the higher the proportion of pro-Islam tweets.
[H2b] The higher the proportion of Muslims in the
population, the lower the proportion of pro-Islam tweets.
Finally, the personal connections we have may contribute
the most to our view of the world. By interacting with others
on a daily basis we negotiate relationships in order to derive
some benefit, and in this process we change ourselves. Interdependence theory is a social exchange theory that postulates that people weigh costs to achieve the greatest benefits out of their relationships (Thibaut and Kelley 1959).
Rewards may come from both similarities and differences in
the dyad, as long as both parties are equally able and willing
to provide rewards for others. Thus, we formulate the last
hypothesis:
[H3] Within mixed Arab/non-Arab networks, users are
likely to tweet similar content to that of their neighborhood.
Related Work
Recently, Twitter and other online media have been utilized
to re-examine longstanding sociological theories. Providing unprecedented scale, and capturing behaviors heretofore unattainable by standard sociological methods, big social data initiated a new field of computational social science (Lazer et al. 2009). Below we describe works on communication and opinion formation most relevant to this paper, and direct the reader to (Mejova, Weber, and Macy
2015) for a comprehensive view of the field.
Analyses of responses to salient political events on Twitter have ranged from Occupy Wall Street protests (Conover
et al. 2013) and same-sex marriage debates (Zhang and
Counts 2015) in US, to, more internationally, Mexican drug
wars (De Choudhury, Monroy-Hernandez, and Mark 2014),
Ferguson unrest (Jackson and Foucault Welles 2015), and
the Arab Spring protests (Bruns, Highfield, and Burgess
2013; Lotan et al. 2011; Wolfsfeld, Segev, and Sheafer
2013). Although Twitter is often associated with social
movements, as Wolfsfeld et al. point out, “politics comes
first” (Wolfsfeld, Segev, and Sheafer 2013), and is followed
by discussion on social media. Due to the international nature of social media, this discussion, Burns et al. state, is
often by the “outsiders looking in” (Bruns, Highfield, and
Burgess 2013). It is these “outsiders” – both in the West and
Middle East – who are the focus of our present work.
Among the theories we consider, Clash of Civilizations
has been revisited by State et al. (State et al. 2015) using Twitter, who found the clusters of countries in the international communication network to resemble the “civilizations” defined by Huntington. Other works on interpersonal interaction, including hashtag usage propagation (Romero, Meeder, and Kleinberg 2011), health behavior (Abbar, Mejova, and Weber 2014), vote turnout (Bond et
al. 2012), and news (Kwak et al. 2010), use immediate user
neighborhood to predict behavior, inadvertently challenging Interdependence theory, wherein social relationships are
Next, we turn to meso-scale dynamics with Density theory, which postulates that individual behavior is “density
dependent and hence varies as a function of aggregate population characteristics” (Huckfeldt 2009b). First applied in
the context of urbanization in 1938, Wirth uses density to
describe the behavioral pressures social heterogeneity puts
on individuals (Wirth 1938). These pressures continue to
be central to the study of opinion and behavior, including as
expressed online (see for example “Bowling alone but tweeting together” (Antoci, Sabatini, and Sodini 2014) or “Online
social networks and trust” (Sabatini and Sarracino 2015)).
Accordingly, the reaction to the Charlie Hebdo attack
does not depend exclusively on individual beliefs or geographic distance, but also on the offline social context in
which the individual is embedded. Concretely, the proportion of people from a different culture surrounding an individual may prompt a shift in one’s beliefs and attitudes. The
prominence of Muslim diasporas in the Western countries
may prompt two possible reactions: (1) the heightened interaction with Muslim population provides a common ground
in Westerners for understanding and empathy, or (2) the ma-
3
#JeSuisAhmed
#JeSuisCharlie
muslim
years, year, old, remembering, outside, attackers, cartoonists, while, guy, shot
jew, christian, frankdeleeuw, merry, jewish, russia,
christmas, customers, jews, zionists
islam
love muhammad, war, isis, wrong, islamicstate, truth,
anti, obama, nd
christianity, judaism, islamism, bible, kkk, religionkills,
atheism, reform, teaches, teachings
freedom
democracy, double, comes, support, liberty, religious,
offensive, insulting, women, without
free, democracy, includes, principle, cornerstone, principles, trumps, limits, essential, speech
press
insulting, values, law, called, liberty, line, double,
democracy, offensive, women
defenders, claiming, speech, slams, censor, while, principle, defence, countries, advocate
terror
protest, since, tomorrow, mosques, pictures, new, tag,
war, wake, days
terrorist, fatah, deadly, chechnya, terrorism, attacks,
savage, senegal, gatestoneist, warns
Table 1: Word associations produced by word2vec for #JeSuisAhmed and #JeSuisCharlie hashtag collections.
negotiated for some mutual benefit. For example, (Abbar,
Mejova, and Weber 2014) model the health value of users
Twitter feed by considering the number of network connections they have who post unhealthy content. Fewer studies
have been done on an intermediate community-level scale.
Community socio-economic well-being has been studied
by (Quercia et al. 2012), who apply sentiment analysis to
tweets from London, and show a significant correlation between the Index of Multiple Deprivation and the “Gross
Community Happiness” score they define. We go a step further, characterizing the mixing of communities, and the effect this mixing has on their opinions, as expressed on Twitter. For both personal as well as larger scales, our work is
a contribution to the ongoing effort to re-examine existing
sociological theories in the sphere of social media.
Recently, hashtags concerning Charlie Hebdo, and specifically “Je Ne Suis Pas Charlie” (“I am not Charlie”), have
been examined by Giglietto & Lee (Giglietto and Lee 2015),
who found a high proportion of retweets and image sharing,
with a unique practice of retweeting nothing but the hashtag
itself (in 2% of the cases). This hashtag, the authors conclude, is a “discursive device that facilitated users to form,
enhance, and strategically declare their self-identity”. In this
work, we attempt to uncover the mechanisms underlying
such self-identification.
tweets by the urban artist banksy. (Similarly, in shorthand
it will be referred to as #JSC.)
#JeNeSuisPasCharlie (I am not Charlie) – May convey two meanings: Rejection of freedom of speech and
freedom of the press when the message is offensive towards Islam. Alternatively, rejection of freedom of speech
for Muslims in Christian countries. It is associated with
prominent reporters as Max Blumenthal and Benjamin
Norton. (#JNSPC)
#JeSuisAhmed (I am Ahmed) – Reactions that differentiate between Islam and terror; emphasis on the fact that
among those defending freedom of speech there are also
Muslims, such as Ahmed, one of the policemen killed
by the terrorists. It is associated with the murdered policeman Ahmed, who was tweeted to be “protecting free
speech” or other french people. Note that this stance is not
necessarily in opposition to #JeSuisCharlie, in fact 76.5%
of those tweeting #JeSuisAhmed also mention #JeSuisCharlie (though only 6.17% do the opposite). (#JSA)
Throughout this project, we focus in particular on #JeSuisCharlie and #JeSuisAhmed – hashtags representing two
distinct positions. The former is a radical defense of freedom of speech; the latter is a defense of the compatibility
between Islam and freedom of speech (though in some cases
limited freedom). These positions are not altogether mutually exclusive, but they do emphasize two different, sometimes opposing, aspects of the same phenomenon.
To take a closer look at the stances associated with these
hashtags, we use word2vec (Mikolov et al. 2013), a computational framework that learns a vector representation of
words by taking a text corpus as input. Table 1 lists the
words associated with a selection of topics. Both #JeSuisCharlie and #JeSuisAhmed hashtags connect freedom with
democracy. Clearly, freedom is understood by all as a democratic value. However, for #JeSuisAhmed users, freedom
is also attached to more negative meanings, such as offense
against Islam, whereas for #JeSuisCharlie users freedom is
treated as an essential principle that should not be trumped
by any other.
Reactions to the Charlie Hebdo attack
Reactions to the Charlie Hebdo attack have clustered around
the following hashtags which we use in our study. We describe each and outline other prominent hashtags associated
with them:
#CharlieHebdo – Sympathy towards the victims of the
attack, general condemnation of the attack. It is associated with informational tags mentioning Paris, the cartoonists killed, and the suspects. (In this paper we will
use a shorthand #CH.)
#JeSuisCharlie (I am Charlie) – Endorsement of freedom of speech and freedom of the press under any circumstances. It is more focused on freedom, and the responses
to the event in form of the tributes, many of them drawings of pens as symbols of writers, and especially popular
4
Data & Methodology
users for #JeSuisCharlie, and 169,598 users for #JeSuisAhmed). Although the volume of tweets is less than the
original, since some of the tweets no longer exist, aggregate
statistics are similar to what Nick Ruest has reported – for
example, in this dataset, 76.74% of tweets are retweets and
1.77% of them are replies, with the most retweeted tweets
having images.
The activity level of the users varies widely across our
dataset – up to the maximum of 35,418 tweets by one user,
with the median of 1 and the mean of 3.69. To remove abnormally active users, who are likely to be spammers, we
discard users who tweeted more than 148 times, which is the
99th percentile of the distribution. This filters out 0.1% users
(2,787) with 9% of total tweets (881,100) from the dataset.
Then to further focus on those users who show their stance
regarding the CharlieHebdo incident strongly and somewhat
unambiguously, we only consider users with two or more
tweets in this dataset, resulting in a dataset of 1,389,673
users with 8,796,872 tweets.
To map the tweets to their respective countries of origin, we geo-locate the data in two ways. First, we look at
whether the tweet is geo-tagged and, if it is, we use it as
user’s location. In a case where a user’s tweets are in different countries, we discard these users to avoid ambiguity. If
geo-tagging is not available, then we apply Yahoo! PlaceMaker4 to the location field in their bio on Twitter. Yahoo!
PlaceMaker is a web service which, given a text, returns best
matched location. For example, with the sentence “I live in
New York”, it returns “New York, New York, USA”. The
service is especially suitable for our data, as it supports languages beyond English.
Among 1,389,673 users, we successfully located 688,651
(45,717 users from geo-tagged tweets and 642,934 Yahoo!
PlaceMaker)5 . These users are mostly from North America and Europe – the top five countries are US, France,
UK, Spain and Canada. Note that we discard users with
two or more locations (e.g., India/Paris, Dubai/Paris) – 221
users when using geo-tagged tweets and 17,352 (2.6%) users
when using Yahoo! PlaceMaker.
Finally, among those located users, 464,176 are in the 39
countries of our interest. In the forthcoming analysis, we focus on these 464,176 users, who are engaged with the Charlie Hebdo shootings, have expressed an opinion on it, and
could be located geographically, along with their 3,030,558
tweets (1.37M of #CH, 1.62M of #JSC, and 42,029 of
#JSA). These users are mostly located in five countries, in
order of magnitude: France, United States, United Kingdom, Spain, and Italy.
We expand this collection by crawling the most recent
tweets (maximum of 3,200) of each of these users. By
detecting mentions in these tweets (handles of other Twitter
users), we then build an ego mention network for the users
in our dataset. We collect 932,003,251 tweets in total and
we extract 23,406,770 mentioned users in those tweets.
Western and Islamic “civilizations”
To compare the Western and Islamic cultures, we focus on
the 39 countries, 20 countries including Western Europe and
the USA, which represent the Western civilizational culture,
and 19 countries from the Middle-East, which represent (not
exhaustively) the Islamic civilizational culture. These countries are listed in Table 2, along with each country’s proportion of Muslim population in parentheses (CIA 2010). The
two groups have a wide difference in the proportion of Muslims, with most Middle Eastern countries having > 70% and
Western < 8%, with notable exceptions such as Cyprus (at
25.3%), which has a distinct population composition.
Region
Country (Muslim population (%))
Middle
East (19)
Morocco (99.9), Iran (99.5), Tunisia (99.5),
Yemen (99.1), Iraq (99), Turkey (98), Algeria
(97.9), Palestine (97.6), Jordan (97.2), Libya
(96.6), Egypt (94.4), Saudi Arabia (93), Syria
(92.8), Oman (85.9), United Arab Emirates
(76.9), Kuwait (74.1), Bahrain (70.3), Qatar
(67.7), Lebanon (61.3)
Western
(20)
Cyprus (25.3), France (7.5), Netherlands (6),
Belgium (5.9), Germany (5.8), Switzerland
(5.5), Austria (5.4), Greece (5.3), Sweden (4.6),
United Kingdom (4.4), Denmark (4.1), Italy
(3.7), Norway (3.7), Luxembourg (2.3), Spain
(2.1), Ireland (1.1), Finland (0.8), Portugal
(0.6), Iceland (0.2), USA (0.9)
Table 2: Selected Middle Eastern and Western countries
(with % Muslim population).
Twitter data
Before focusing on the individuals within countries, however, we collect tweets concerning the Charlie Hebdo
incident using two sources: (1) Nick Ruest’s collection of
tweets which track #JeSuisCharlie, #JeSuisAhmed, and
#CharlieHebdo, and (2) a Topsy.com collection tracking
#JeNeSuisPasCharlie and #JeSuisPasCharlie.
Nick Ruest collection. We use a collection created by
Nick Ruest2 , who has collected tweets that include one
of the following three hashtags – #JeSuisCharlie, #JeSuisAhmed, and #CharlieHebdo – from 2015-01-07 11:59:12
UTC to 2015-01-28 18:15:35 UTC using Twitter’s search
API.
We “hydrated” (i.e. collected metadata for) the released
tweet IDs3 using Twitter public API, collecting 11,367,987
tweets (7.1M tweets with #CharlieHebdo, 6.5M with #JeSuisCharlie, and 264,097 with #JeSuisAhmed) posted by
3,081,039 unique users (2M users for #CharlieHebdo, 2M
4
https://developer.yahoo.com/boss/geo/
We have 13,823 users who got located by both Geo-tagged
tweets and Yahoo PlaceMaker. For the 92.3% users (12,756), two
methods are resulted in the same location.
2
5
http://goo.gl/fI0QPU
3
http://dataverse.scholarsportal.info/dvn/dv/nruest/faces/study/
StudyPage.xhtml?globalId=hdl:10864/10830
5
We attempt to locate these mentioned users using their
geo-tagged tweets and self-described location (as above)
and successfully find 4,326,045 users’ location (18.4%).
Among 274,152,345 links between our seeding users to
mentioned users, 0.6% (1,779,086) of them are reciprocal
links between users who tweeted CH.
Hashtag
30
●
CH
JNSPC
JSA
JSC
Total
Volume
Normalized Volume (%)
8e+05
●
●
20
6e+05
4e+05
●
●
●
2e+05
●
●
●
0e+00
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
06 07 08 09 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29
January 2015
10
Topsy collection. We collect the tweets containing one
of the two hashtags not available in the Nick Ruest collection – #JeNeSuisPasCharlie (JNSPC) and #JeSuisPasCharlie
(JSPC), both versions of “I am not Charlie” – using Topsy6
from 7th to 28th January 2015. Topsy is a certified partner of Twitter for offering social search and social analytics,
such as Twitter Oscar Index7 and Twitter political Index8 .
Topsy indexes every public tweet and allows users to search
them by certain keywords since 20139 . This means that our
analysis is based on the entire set of public tweets instead
of small-sized samples. While Tospy offers the public interface to access to tweets even after it was acquired by Apple
in 2013, Apple finally shutdowns the service as of December
2015.
We initially gather 35,966 tweets (tweet id, screen name
of users, and text) from Topsy. Then using Twitter API, we
collect 32,315 tweets (30,638 (JNSPC) and 5,379 (JSPC))
with 21,276 users. We then filter out users who have high
activity level (512) and users who have only one tweet
(16,919). Among the 4,356 remaining users, 395 users are
live in one of 39 countries of our interest. We focus on these
395 users and their 1,404 tweets for the analysis. We then
crawl 945,762 recent tweets posted by these 395 users. Out
of 159,028 users mentioned in those tweets, 12.29% of users
(19,529) are located.
These tweets are coming from locations that are somewhat different from our previous dataset. The top 5 countries
where these users are located are France, Algeria, United
States, Morocco, and Belgium.
The normalized temporal volume (showing percentage of
total hashtag volume) of the final collection (after user geolocation and selection) can be found in Figure 1, and a raw
volume can be found as an inset plot. The vast majority
of activity happens within 3 days of the event, with #CharlieHebdo dominating the volume. The use of #JeSuisAhmed
peaks on the day after the attack. //
Arabic identification. As our hypotheses deal with
users’ religious identities, we need to differentiate the Muslims from the non-Muslims among the users in our dataset.
Since Twitter users usually do not declare their religious
identities in their profiles, we proceed with the – admittedly
rough – assumption that Arabic speakers, or users with Arabic names, are Muslim. All other names or languages are
non-Muslim. Considering that approximately 94% of Arabs
are Muslims (CIA 2010), the assumption can be reasonably
accepted. Also, it is worth to mention that Iran and Turkey
●
●
●
●
●
12
13
●
●
●
16
17
0
06
07
08
09
10
11
14
15
●
●
●
●
●
●
●
●
●
●
●
18
19
20
21
22
23
24
25
26
27
28
29
January 2015
Figure 1: Daily tweet volume mentioning each of four hashtags (#CH, #JSC, #JSA, and #JNSPC). Insert: original data
without user selection.
are Muslim countries (99.5% and 98% of populations are
Muslims, respectively, as in Table 2), but they are non-Arab
countries. We thus exclude users from Turkey and Iran to
avoid bias in the experiments using Arab and non-Arab distinction.
We firstly detect any user who tweets in Arabic, has a
name in Arabic, or set their language on Twitter as Arabic.
To detect the language of each tweet, we use three widelyused libraries for language detection, which are CLD2 (embedded in Google Chrome)10 , langid.py (Lui and Baldwin
2012), and LangDetect11 , and mark language by simple majority voting. It is known that this ensemble approach consistently outperforms any individual system, including Twitter’s language metadata (VRL 2014). If the name is not in
Arabic, then we check it against a dictionary of 4,401 Arabic
names in English (2,160 male names, 2,151 female names,
and 100 neutral names), which we build using baby name
lexicons12 . The list of names used in the analysis is available at 13 .
In our seeding dataset, we find that 5.3% of users (23,924)
pass the above filters. Among them, 69.8% of users (16,705)
are detected by the name-based approach, while 27.0% of
users (6,469) are detected by their language use. Only 750
users are detected by both methods.
For the rest of the paper, we will use Arab/non-Arab
distinction for the users identified via the above method, not
to confuse it with other sources of religious identity (such
as that identified by CIA Fact Book and listed in Table 2).
Language. The languages used in our collections are
shown in Table 3. For Non-Arab users, French is the
most used language at 47.93% of all tweets with English at
35.57%. For Arab users, English is the most used language
at 49.57% of all tweets, with French at 25.85%, and Arabic
at only 9.89%. The latter statistic is understandable, since
10
http://blog.mikemccandless.com/2011/10/accuracy-andperformance-of-googles.html
11
https://github.com/shuyo/language-detection/blob/wiki/
ProjectHome.md
12
http://www.searchtruth.com/baby nameshttp://www.
urduseek.com/names
13
https://goo.gl/Nam1ts
6
http://topsy.com/
http://oscars.topsy.com/
8
https://election.twitter.com/
9
http://about.topsy.com/2013/09/04/every-tweet-everpublished-now-at-your-fingertips
7
6
Hashtag
Language
CharlieHebdo
Arabic
English
French
Others
Arabic
English
French
Others
Arabic
English
French
Others
Arabic
English
French
Others
Arabic
English
French
Others
JeSuisCharlie
JeSuisAhmed
JNSPC
Total
Arab
Non-Arab
Hashtag
Non-arab
Arab
Western
Middle east
11,846 (13.74%)
44,316 (51.42%)
17,479 (20.28%)
12,551 (14.56%)
2,008 (3.94%)
22,531 (44.16%)
18,428 (36.11%)
8,060 (15.80%)
207 (4.11%)
3,702 (73.48%)
824 (16.36%)
305 (6.05%)
41 (15.77%)
98 (37.69%)
111 (42.69%)
10 (3.85%)
14,102 (9.89%)
70,647 (49.57%)
36,842 (25.85%)
20,926 (14.68%)
0 (0.00%)
513,918 (40.45%)
493,908 (38.87%)
262,750 (20.68%)
0 (0.00%)
461,233 (30.98%)
832,313 (55.90%)
195,314 (13.12%)
0 (0.00%)
18,929 (53.66%)
13,299 (37.70%)
3,050 (8.65%)
0 (0.00%)
272 (29.73%)
547 (59.78%)
96 (10.49%)
0 (0.00%)
994,352 (35.57%)
1,340,067 (47.93%)
461,210 (16.50%)
JSC
JSA
JNSPC
97.67
2.27
0.06
88.09
10.88
1.03
97.65
2.29
0.07
90.51
8.94
0.07
Table 4: Percentage of tweets mentioning each hashtag.
groups use the largely topical #CharlieHebdo hashtag, and
very little #JeNeSuisPasCharlie. However, the relative proportion of #JeSuisCharlie to #JeSuisAhmed is strikingly different, with one #JeSuisAhmed to every 10 #JeSuisCharlie
for the Arab users, and one to 43 for non-Arab ones. Similar
distinction is evident when we segment users by geographical locations (Western vs. Middle East).
Thus, we find some support for H1, although both populations use #JeSuisCharlie more than #JeSuisAhmed, and
this cannot be explained by the Clash of Civilizations theory.
The wordclouds in Figure 2 show how Non-Arab and Arab
users use #JeSuisCharlie, with Arabs mentioning Ahmed,
God, and solidarity while both focusing on freedom.
Table 3: The fraction of tweets in different languages by
Arabs and non-Arabs. (Classified using pooled language detection.)
the queries were made using French hashtags, and in latin
alphabet, which surely excluded those tweets written purely
in Arabic (for more on this limitation, see the Discussion
section).
The large amount of users classified as Arabs that use languages other than Arabic is also probably due to the fact
that the largest number of tweets are concentrated in countries like France, United Kingdom, and the USA. This means
that a lot of users with an Arab background living in these
countries are tweeting in English and French, not Arabic.
(a) #JeSuisCharlie by NonArab
(b) #JeSuisCharlie by Arab
Figure 2: Wordclouds for #JeSuisCharlie collection by NonArab vs. Arab users.
Results
In this section we present our findings regarding the three
posed theories modeling the formation of opinions expressed wherein.
Density theory
Density theory claims that the population densities of culturally diverse groups in the individual’s offline social context are important factors in the formation of opinion. In
this case, population densities are characterized by the size
of groups sharing the same “civilizational” culture within
one country. Is it possible that a diaspora of Arabs in the
West, and Westerners in the Arab world, affects the understanding of and attitudes toward Charlie Hebdo event?
Such effects may be simultaneous and contradictory: on one
side, they could be promoting empathy and understanding
by co-habitation, on another, they could encourage hostility
to an increasingly visible minority (from the point of view
of Muslims in the Middle East or Westerners in the West)
or towards an unfriendly majority (from the point of view of
Muslims in Western countries or Westerners in the Middle
East). Two alternatives arise in the face of minority/majority
interactions (here, for Western users):
Clash of civilizations theory
Under Huntington’s thesis, the major fault lines in post-Cold
War geo-politics lie along cultural and religious identities.
In this study, the users we consider can be roughly divided
as belonging to two “civilizations” – the Western Christian
civilization and the Islamic civilization. Huntington poses
that Muslims, by virtue of belonging to the Islamic civilization, will be more wary of defending freedom of speech than
Westerners. Here we test this hypothesis.
[H1] Opinions expressed about the Charlie Hebdo shootings are divided along “civilizational” faultlines, with a
higher proportion of pro-free speech tweets by users in Western Christian civilization countries, and a higher one of proMuslim tweets by users in the Islamic civilization countries.
Table 4 shows the proportion in the use of hashtags by
users identified as Arab and all others (Non-Arab). Both
7
50
50
50
Libya
Libya
●
30
40
40
30
Egypt
Jordan
● ●
20
Qatar
●Saudi Arabia
Arab Emirates
●United
●Kuwait
●Oman
●Algeria
●Morocco
●Palestine
●Bahrain
●Yemen
●Iraq
●Tunisia
●Syria
●Turkey
●Lebanon
●Iran
●
10
Norway
Sweden
United
Kingdom
Denmark
United
States
Ireland
Finland
Austria
NetherlandsCyprus
Germany
Switzerland
Belgium
Italy
Greece
Iceland
Portugal
Spain
Luxembourg
France
30
20
Bahrain
●
●Egypt
●Saudi Arabia
●Jordan
●Libya
●Oman
●Morocco
●Kuwait
●Algeria
Emirates
Qatar●United Arab●Palestine
Yemen
●
●
10
Tunisia
Iraq
●●
Syria
●
Norway
Sweden
United
Kingdom
Denmark
United
States
Ireland
Finland
Austria
NetherlandsCyprus
Germany
Switzerland
Italy
Belgium
Greece
Iceland
Spain
Portugal
France
Luxembourg
0
Percentage of JSA tweets (%)
40
Percentage of JSA tweets (%)
Percentage of JSA tweets (%)
●
25
50
75
100
Percentage of Muslim in the country (%)
25
10
Lebanon
75
100
Norway
United Kingdom
Denmark
United
States
Ireland
Sweden
Austria
Belgium
Luxembourg
Switzerland
Germany
Netherlands
Greece
Portugal
Italy
France
Spain
Finland
Cyprus
Iceland
0
Percentage of Muslim in the country (%)
(a) All users
25
Bahrain
●
Syria
●
Lebanon
●
50
75
100
Percentage of Muslim in the country (%)
(b) Non-Arab users
15
Arabia
Oman
●Saudi
●Iraq
●Morocco
●Yemen
●Tunisia
●
●
50
Kuwait
●
0
0
●Algeria
●Jordan
Egypt
●Palestine
●
20
0
0
●United Arab Emirates
Qatar
●
(c) Arab users
15
15
10
Norway
Sweden
5
United States
Ireland
Finland
Iceland
Portugal
United Kingdom
Denmark
Austria
Netherlands
Germany
Switzerland
Belgium
Greece
France
Italy
Spain
Luxembourg
0
2.5
5.0
7.5
Percentage of Muslim in the country (%)
(d) All users (western)
Denmark
United States
Ireland
10
Norway
Sweden
5
United Kingdom
Denmark
United States
Ireland
Finland
Iceland
Portugal
Austria
Netherlands
Germany
Switzerland
Belgium
Greece
France
Italy
Spain
Luxembourg
0
0.0
Percentage of JSA tweets (%)
10
Percentage of JSA tweets (%)
Percentage of JSA tweets (%)
Norway
United Kingdom
Sweden
Austria
5
2.5
5.0
7.5
Percentage of Muslim in the country (%)
(e) Non-Arab users (western)
Netherlands
Greece
Portugal
Italy
France
Spain
Finland
0
0.0
Belgium
Switzerland
Germany
Luxembourg
Iceland
0.0
2.5
5.0
7.5
Percentage of Muslim in the country (%)
(f) Arab users (western)
Figure 3: The percentage of JSA tweets over JSA+JSC tweets by Muslim population of the country, comparing 3 different
groups: all users, non-Arab, and Arab. Arab countries are colored in red and non-Arab in blue. Due to Arab filter design,
Turkey and Iran are removed from figures b, c, e, and f.
[H2a] The higher the proportion of Muslims in the population, the higher the proportion of pro-Islam tweets.
[H2b] The higher the proportion of Muslims in the population, the lower the proportion of pro-Islam tweets.
Mirror hypotheses can be posed for Arab users. However,
our conclusions are more sound for Western population due
to the languages of our dataset, so we focus on this group of
users.
Here, we take advantage of The World Factbook’s proportion of Muslim residents, as described in Data Section. Figure 3(a) plots the percent of #JeSuisAhmed (#JSA) tweets
over the combined total of #JeSuisAhmed and #JeSuisCharlie (y-axis) against the proportion of Muslims in the country
(x-axis). Figure 3(d) shows a zoom of the bottom left corner
of Figure 3(a), where Western countries are clustered (except Cyprus, which has 25.3% Muslim population).
To compare the behavior of Arab and non-Arab users (as
defined in Data Section), we present the two user populations in Figures 3(b,e) for non-Arab users and Figures 3(c,f)
for Arab ones. In these graphs, we exclude Turkey and Iran
to eliminate bias, as users from these countries are not Arabs
but are Muslims nevertheless.
Table 5 shows Pearson product-moment correlation r and
Spearman rank correlation coefficient ρ between the percentage of #JSA tweets and the percentage of Muslims in the
country’s population in various slices of data. As Figure 3(b)
shows, there is a clear positive correlation (Pearson r=0.845,
p < 0.001), suggesting that Westerners who live in Middle
Eastern countries tend to tweet more with #JSA than those
who live in the West. There is, therefore, a clustered division along the two “civilizations” described by Huntington.
However, the story is more complicated when we go deeper
and pay attention to the social context.
According to the Clash of Civilizations theory, non-Arabs
(i.e. Westerners) living in the Middle-East should behave in
a similar way to non-Arabs living in the West; after all, they
are all non-Muslims and they belong to the Western “civilizational” culture. Figure 3(b), according to Huntington,
should show all countries clustered on the left bottom corner. The graph shows, on the contrary, that non-Arabs living
in the Middle East, where they are surrounded by large majorities of Muslims, are much more likely to use #JSA than
8
All countries
All users
Non-Arab
Arab
Western
and in the case they are Arab, increased awareness of the
Arab point of view.
We divide users into two groups: (1) users who have not
mentioned any Arabs in their tweets at all (28,939, denoted
as “No Mentions”) and (2) users who have mentioned an
Arab user at least once (338,430, denoted as “Some mentions”). We then compare the use of #JeSuisAhmed between
the groups, and find that the mixed group uses #JSA more
than twice as much as the homogeneously non-Arab group,
with 3.61% compared to 1.31% likelihood, respectively. A
Welch’s t test confirms that the difference in two groups is
statistically significant (t44,164 = 38.80, p < 0.001).
Arab
Person (r)
Spear. (ρ)
r
ρ
r
ρ
0.745***
0.845***
0.675***
0.698***
0.740***
0.675***
-0.004
0.021
-0.186
0.136
0.130
0.097
0.064
0.193
0.157
-0.300
-0.010
-0.022
Significance: p <0.0001 ***, p < 0.001 **, p < 0.01 *
Table 5: Pearson and Spearman correlations of % of JSA
tweets to the % of Muslims in the country.
Percentage of JSA tweets (%)
non-Arabs living in the West. A similar observation can be
made for Arabs in the West (which all should cluster at the
top right of Figure 3(c), but do not).
If we now turn to users living in the West, we also see that
the density of the social context matters. For both non-Arabs
and Arabs the correlation is extremely weak (see “Western”
column of Table 5). However, Figure 3(d) seems to suggest
that the relationship between the number of #JSA hashtags
and the percentage of Muslims in the country might not be
linear, but concave downwards. At between 0 and 3.5% of
Muslims in the country, non-Arabs are more likely to use
#JSA the larger the number of Muslims that live in the country; after a tipping point of 3.5% of Muslims in the country,
however, non-Arabs are less likely to hashtag JSA the larger
the number of Muslims surrounding them. Therefore, the
Muslim minority helps non-Muslims to be more emphatic
as far as this minority is not too large. The tipping point at
which non-Arabs become less emphatic and more fearful of
the Arab point of view is approximately at 3.5% of Muslim
population. Italy would seem to be the only clear outlier of
this concave relationship.
To verify the robustness of these figures, we model this
behavior using a measure of religiosity (indication of how
important religion is to a country’s residents). Indeed, religiosity, as measured by Gallup in 200914 , is highly correlated with the proportion of #JSA tweets at r = 0.7085.
However, when a linear regression is fitted using both religiosity and rate of Muslim population, the effect of religiosity is lost.
15
● Arab in Arab countries
Arab in Non−Arab countries
Non−Arab in Arab countries
Non−Arab in Non−Arab countries
●
10
5
0
No Arab mention
>= 1 Arab mention
Figure 4: Mean percentage of JSA tweets for four user
groups in conditions with and without Arab mentions.
To understand better whether the mention network effect
is confounded by any offline effect, such as country of residence, we now look at four different user groups: a) Arabs
living in Arab countries, b) Arabs living in Non-Arab countries, c) Non-Arabs living in Arab countries, and d) NonArabs living in Non-Arab countries and examine to what
extent the online factor plays a role. In Figure 4 we show
the behavior of each group. Mention network factor plays
a role for all user groups except Arabs in Arab countries,
which due to sparsity we do not consider (there are only 24
users, all of whom mention some Arab users).
Since the majority of users is in the “non-Arabs in nonArab countries” group, the result is similar to what we observe when we consider all users (see earlier paragraph).
The means of No Mention and Some Mentions are 1.26 and
3.03, respectively (t42,018 = 30.19, p < 0.001). The next
strongest relationship is for “Arabs in non-Arab countries”
user group, with the likelihood of tweeting #JSA almost
doubling from 5.37 to 9.86 (t320 = 3.62, p < 0.001). The last
group, “non-Arabs in Arab countries” also shows a strong
pattern, with the difference between 2.83 and 15.76 having
p < 0.001 (t72 = 6.56).
Overall, we observe that the personal mentions in users’
interactions do affect the likelihood of expressing an opinion favorable to #JeSuisAhmed. Note, however, that the “offline” distinction – that is, where the user lives – is a stronger
predictor of online behavior.
Among the 27.27M users, including the mentioned users,
our filters detect 4.8% (1,312,008) Arab users. We find that
the links in mention network are mostly to non-Arab Twitter
users – 90.07% links from non-Arabs and 5.54% links from
Interdependence theory
Whereas density theory concerns the aggregate level of
countries, we now turn to the individual level of analysis, in
which individuals build interpersonal relationships which affect both parties. Interdependence theory concerns the effect
of online interactions on the individual’s online behavior:
[H3] Within mixed Arab/non-Arab networks, users are
likely to tweet similar content to that of their neighborhood.
As mentioned in the Data & Methodology Section, we
build a mention network for each user in our dataset. This
network contains all users whose Twitter handles have been
mentioned in the tweets of our users. Those users were then
also labeled as Arab or not. These mentions signify a user’s
connection to, or at least awareness of, other Twitter users,
14
http://www.gallup.com/poll/142727/religiosity-highestworld-poorest-nations.aspx
9
All users
Non-Arab users
Non-Arab in Non-Arab countries
Non-Arab in Arab countries
Arab users
Arab in Arab countries
Arab in Non-Arab countries
Person (p)
Spearman (r)
0.215***
0.208***
0.171***
0.134***
0.156***
0.134***
0.171***
0.153***
0.130***
0.118***
0.121***
0.191***
0.122***
0.118***
identification, thus, are important for successful online communities.
As we mentioned earlier, scholars have long worked on
understanding social responses to political events, especially on social media (Bruns, Highfield, and Burgess 2013;
Conover et al. 2013; Zhang and Counts 2015). Our work is
aligned with the study by Burns et al. in that it takes into
account the global characteristics of social media around
the Arab Spring (Bruns, Highfield, and Burgess 2013). As
their study shows information flow between Arabic and nonArabic user groups by looking into reply and retweets, our
work illustrates, instead, how such data could be used for
international-scale verification of existing hypotheses developed in social and political science.
Analyses described in this work, as in most social behavior studies, must be interpreted within correlation is not
causation warning. The captured phenomena is likely in
part due to homophily, wherein more tolerant people would
connect to a more diverse sphere of friends. The next step,
then, is to simulate, or indeed perform, experimental evaluation in order to verify the causal links between interaction
with diverse communities and opinion change. Social media
giants such as Facebook and Twitter are in a unique opportunity to monitor the readership behaviors of their users, however a strict adherence to privacy and non-manipulation considerations must be implemented (for such studies as (Bakshy, Messing, and Adamic 2015), for example).
Finally, the role of mass media may play a central stage in
the opinion formation and propagation in social media – an
important dimension for future study. Moreover, as (Lin and
Margolin 2014) show, in social media attention tends to converge on few hashtags which signal the topic. Thus, more
fine-grained topic analysis may find stances in line with #JeSuisAhmed in the #JeSuisCharlie stream.
Significance: p <0.0001 ***, p < 0.001 **, p < 0.01 *
Table 6: Pearson and Spearman correlations of % of JSA
tweets to % of Arab mentions in the mention network by
different user groups.
Arabs. Only 3.73% links mention Arab users (2.67% from
non-Arabs and 1.06% from Arabs). Thus the discussion in
our dataset is focused on the Western world.
Table 6 shows how the percentage of Arab mentions in
one’s mention network is associated with the percentage of
JSA tweets. We find a positive relationship across all different user groups, weak but statistically significant.
Discussion
The results of this study must be seen in the light of two
technical limitations, both of which would serve as important future directions of research.
The data we have considered here has been collected
using French hashtags, and in Latin alphabet. Although
many other languages, including English, were captured,
this method has surely missed relevant Arabic content. Capturing the multilingual response to international news is an
important technical challenge for the worldwide opinion
tracking community. Another challenge is the identification of religious affiliation purely from online data. Automatic classification, such as the one proposed by Nguyen &
Lim (Nguyen and Lim 2014), may provide access to users
whose religion does not statistically follow from their name
or language, as we have assumed in this research.
Above limitations aside, the insights in this study have
several implications for human-centric application design.
While it has been studied extensively in the political context,
our study is the first which empirically shows that exposure
to other views affects user behavior in the cultural context.
Diversity is one of the key elements for a healthy society, yet
there is much polarization in both online and offline worlds –
with echo chambers limiting the views of both sides (Gilbert,
Bergstrom, and Karahalios 2009). Our findings support the
design of more pluralistic discourse efforts.
As noted by (Giglietto and Lee 2015), “Je Suis...” hashtags aid users in self-identification as a part of a group.
This kind of behavior has been reported in various contexts (Chen, Sun, and Hsieh 2008). For instance, in online games, guild (small group) members explicitly show
their guild names in their handle names (Nardi and Harris
2006). In a virtual world, expressing oneself and having
a group membership is vital to sustain online communities
and offer better user experience. The affordances for self-
Conclusion
Our work presents a systematic application of sociological
opinion formation theories to the analysis of the Twitter response to the Charlie Hebdo shootings of January 2015. The
theory of the Clash of Civilizations first seemed to be confirmed at face value by the data, but when we look deeper,
paying attention to the social context (i.e. the country and
its socio-demographic composition) and the structure of online interactions between users (culturally mixed or culturally homogeneous), we see that Clash of Civilizations needs
to be rejected, or at least qualified, in favor of Density theory and Interdependence theories. Culture – and religion as
a fundamental part of it – matters a great deal, as Huntington
argues, but it matters in much more subtle ways than those
advanced by the Clash of Civilizations theory.
Social media data makes it possible to model an individual’s interaction with both mainstream and minority cultures, allowing us to model individual behavior change. As
geo-political developments unfold, and greater number of
cultures will come in contact, this data will increasingly
present opportunities for verifying old and forming new theories on opinion formation in pluralistic societies.
10
References
Gutmann, M.; Jebara, T.; King, G.; Macy, M.; Roy, D.; and
Van Alstyne, M. 2009. Computational social science. Science
323(5915):721–723.
Lin, Y.-R., and Margolin, D. 2014. The ripple of fear, sympathy and solidarity during the boston bombings. EPJ Data
Science 3(1):1–28.
Lotan, G.; Graeff, E.; Ananny, M.; Gaffney, D.; Pearce, I.;
et al. 2011. The arab spring. the revolutions were tweeted:
Information flows during the 2011 tunisian and egyptian revolutions. International journal of communication 5:31.
Lui, M., and Baldwin, T. 2012. langid. py: An off-the-shelf
language identification tool. In Proceedings of the ACL 2012
system demonstrations, 25–30. Association for Computational
Linguistics.
Mejova, Y.; Weber, I.; and Macy, M. W. 2015. Twitter: A
Digital Socioscope. Cambridge University Press.
Mikolov, T.; Sutskever, I.; Chen, K.; Corrado, G. S.; and Dean,
J. 2013. Distributed representations of words and phrases
and their compositionality. In Advances in neural information
processing systems, 3111–3119.
Nardi, B., and Harris, J. 2006. Strangers and friends: Collaborative play in world of warcraft. In CSCW, 149–158. ACM.
Nguyen, M.-T., and Lim, E.-P. 2014. On predicting religion
labels in microblogging networks. In Proceedings of the 37th
international ACM SIGIR conference on Research & development in information retrieval, 1211–1214. ACM.
Przeworski, A. 1974. Contextual models of political behaviour. In Political Methodology, volume 1. 27–61.
Quercia, D.; Ellis, J.; Capra, L.; and Crowcroft, J. 2012.
Tracking gross community happiness from tweets. In Proceedings of the ACM 2012 conference on Computer Supported
Cooperative Work, 965–968. ACM.
Romero, D. M.; Meeder, B.; and Kleinberg, J. 2011. Differences in the mechanics of information diffusion across topics:
idioms, political hashtags, and complex contagion on twitter.
In WWW, 695–704. ACM.
Sabatini, F., and Sarracino, F. 2015. Online social networks
and trust. Munich Personal RePEc Archive.
State, B.; Park, P.; Weber, I.; and Macy, M. 2015. The mesh of
civilizations in the global network of digital communication.
PLoS ONE.
Thibaut, J. W., and Kelley, H. H. 1959. The social psychology
of groups.
VRL, N. 2014. Accurate language identification of twitter
messages. In Proceedings of the 5th Workshop on Language
Analysis for Social Media (LASM)@ EACL, 17–25.
Wirth, L. 1938. Urbanism as a way of life. American journal
of sociology 1–24.
Wolfsfeld, G.; Segev, E.; and Sheafer, T. 2013. Social media and the arab spring politics comes first. The International
Journal of Press/Politics 18(2):115–137.
Wright, G. C. 1976. Community structure and voter decision
making in the south. In Public Opinion Quarterly, number 40.
201–215.
Zhang, A. X., and Counts, S. 2015. Modeling ideology and
predicting policy change with social media: Case of same-sex
marriage. In SIGCHI.
Abbar, S.; Mejova, Y.; and Weber, I. 2014. You tweet what
you eat: Studying food consumption through twitter. SIGCHI.
Allardt, E., and Pesonen, P. 1967. Cleavages in finnish politics. In Lipset, S., and Rokkan, S., eds., Party Systems and
Voter Alignments. New York: Free Press.
Antoci, A.; Sabatini, F.; and Sodini, M. 2014. Bowling alone
but tweeting together: the evolution of human interaction in
the social networking era. Quality & Quantity 48(4):1911–
1927.
Bakshy, E.; Messing, S.; and Adamic, L. 2015. Exposure to
diverse information on facebook. Facebook Research Blog.
Bond, R. M.; Fariss, C. J.; Jones, J. J.; Kramer, A. D.; Marlow,
C.; Settle, J. E.; and Fowler, J. H. 2012. A 61-million-person
experiment in social influence and political mobilization. Nature 489(7415):295–298.
Bruns, A.; Highfield, T.; and Burgess, J. 2013. The arab spring
and social media audiences english and arabic twitter users
and their networks. American Behavioral Scientist 57(7):871–
898.
Chen, C.-H.; Sun, C.-T.; and Hsieh, J. 2008. Player guild dynamics and evolution in massively multiplayer online games.
CyberPsychology & Behavior 11(3):293–301.
CIA, E. 2010. The world factbook 2010. Central Intelligence
Agency, Washington, DC.
Conover, M. D.; Ferrara, E.; Menczer, F.; and Flammini, A.
2013. The digital evolution of occupy wall street.
De Choudhury, M.; Monroy-Hernandez, A.; and Mark, G.
2014. Narco emotions: affect and desensitization in social
media during the mexican drug war. In SIGCHI Conference
on Human Factors in Computing Systems. ACM.
Giglietto, F., and Lee, Y. 2015. To be or not to be charlie:
Twitter hashtags as a discourse and counter-discourse in the
aftermath of the 2015 charlie hebdo shooting in france. Workshop on Making Sense of Microposts at the 24th International
World Wide Web Conference.
Gilbert, E.; Bergstrom, T.; and Karahalios, K. 2009. Blogs are
echo chambers: Blogs are echo chambers. In System Sciences,
1–10. IEEE.
Huckfeldt, R. 2009a. Citizenship in democratic politics: Density dependence and the micro-macro divide. In Comparative
Politics: Rationality, Culture, and Structure. New York: Cambridge University Press. 291–313.
Huckfeldt, R. 2009b. Interdependence, density dependence,
and networks in politics. In American Politics Research, volume 37. 921–950.
Huntington, S. P., et al. 1993. The clash of civilizations?
Jackson, S. J., and Foucault Welles, B. 2015. # ferguson
is everywhere: initiators in emerging counterpublic networks.
Information, Communication & Society 1–22.
Kwak, H.; Lee, C.; Park, H.; and Moon, S. 2010. What is
twitter, a social network or a news media? In Proceedings
of the 19th international conference on World wide web, 591–
600. ACM.
Lazer, D.; Pentland, A.; Adamic, L.; Aral, S.; Barabási, A.L.; Brewer, D.; Christakis, N.; Contractor, N.; Fowler, J.;
11